<b>Why almost every anchor "ranking factor" study is correlational — and what that costs you</b>

<b>Why almost every anchor "ranking factor" study is correlational — and what that costs you</b>

Field note on methodology, because it changes how you should read every number in this space.

The dominant study design: collect SERPs, extract anchor distributions of ranking URLs, compute correlations between anchor features and position. The output looks authoritative — "exact-match anchors correlate at r = 0.12 with ranking."

The problem is structural, not fixable with a bigger sample. Anchor features are entangled with everything: strong domains attract both more links <i>and</i> better-optimized anchors; topical pages earn both relevant anchors <i>and</i> relevant content. A correlation between anchors and rank could be anchors causing rank, rank-worthiness causing both, or pure confounding through domain strength.

On one hand, weak positive correlations for descriptive anchors are stable enough across studies to suggest something real underneath. On the other, an r near 0.1 explains ~1% of variance — statistically detectable, practically tiny, and swamped by content and authority.

The only clean design — randomly assign anchors to otherwise-identical pages and watch — is unethical and impossible at scale, so we're stuck with observation.

Practical read: treat anchor correlations as weak priors, never as levers with known gain.

Open question: what's the realistic upper bound on anchor-text's independent causal contribution once domain strength is properly controlled — 1%? 5%? We genuinely don't know.